[coreboot] r3259 - trunk/coreboot-v2/targets/amd/serengeti_cheetah_fam10

[Marc Jones]

Carl-Daniel Hailfinger wrote:
> On 24.04.2008 18:37, Marc Jones wrote:
>   
>> Carl-Daniel Hailfinger wrote:
>>   
>>     
>>> I think most of the size problems can be fixed by removing all those
>>> "inline" keywords (unless this is required by the processor).
>>>
>>> So far, these size problems have only been appearing on Stefan's build
>>> host (probably due to gcc version), so I'm a bit reluctant to see these
>>> patches as more than a band-aid until we remove unnecessary "inline"s.
>>>
>>>
>>>     
>>>       
>> I think some of those inlines should go away. They are certainly left 
>> over from previous versions of the code. They are no longer important 
>> when the ROM is instruction cached and you have CAR for a stack. 
>> Historically, without instruction caching you can loose performance 
>> doing random reads to the ROM. You can also lose any prefetch advantage 
>> your CPU and/or chipset might be doing.
>>   
>>     
>
> Dumb question: Do we use instruction caching during CAR? (I ask because
> I want to make sure the v3 design I have in mind for K8 bringup fits
> reality.)
>
> Regards,
> Carl-Daniel
>
>   
That is what XIP_ROM_BASE  should be setting up. We also turned it on 
for Geode in v3 because the LAR parsing and decompress was taking so long.
Marc


-- 
Marc Jones
Senior Firmware Engineer
(970) 226-9684 Office
mailto:Marc.Jones at amd.com
http://www.amd.com/embeddedprocessors